摘要
为了提高短期电力负荷预测精度,针对负荷数据时序性与非线性的特点,提出一种基于贝叶斯优化的CNNGRU短期电力负荷预测模型。首先,将电力负荷数据按时间滑动窗口构造连续特征作为输入,采用CNN对负荷数据进行特征提取,将特征以时序序列方式作为GRU网络输入;然后通过GRU网络进行短期负荷预测,构建CNN-GRU预测模型。针对CNN-GRU模型易陷入局部最优以及超参数寻找难的问题,利用贝叶斯优化寻找最优超参数组合,对模型进行超参数优化,构建贝叶斯优化的CNN-GRU短期电力负荷预测模型。实验结果表明,贝叶斯优化的CNN-GRU模型的MAE值比传统的CNN-GRU网络模型降低58%,精度提升1.23%,说明所提模型能够有效提高负荷预测精度,可作为短期电力负荷预测工具。
In order to improve the accuracy of short⁃term power load forecasting,a CNN⁃GRU(convolutional neural network⁃gated recurrent unit)short⁃term power load forecasting model based on Bayesian optimization is proposed for the characteristics of timeliness and nonlinearity of load data.Continuous features of power load data are constructed according to the time sliding window as input,and CNN is used to extract features from the load data,and the features are inputted into the GRU network in a time series manner.The short⁃term load forecasting is conducted by the GRU network,and the CNN⁃GRU forecasting model is constructed.In allusion to the problems of CNN⁃GRU model easily falling into local optimization and difficulty in finding hyperparameters,Bayesian optimization is used to find the optimal combination of hyperparameters,optimize the model with hyperparameters,and construct the optimized Bayesian CNN⁃GRU short⁃term power load forecasting model.The experimental results show that the MAE value of the optimized Bayesian CNN⁃GRU network model is reduced by 58%compared with the traditional CNN⁃GRU network model,and the accuracy is improved by 1.23%.The proposed model can effectively improve the accuracy of load forecasting,and can be used as a short⁃term power load forecasting tool.
作者
吴永洪
张智斌
WU Yonghong;ZHANG Zhibin(School of Information Engineering and Automation,Kunming University of Technology,Kunming 650504,China)
出处
《现代电子技术》
2023年第20期125-129,共5页
Modern Electronics Technique
基金
2020年第一批贵州省职业教育兴黔富民行动计划立项精品课程建设项目。
作者简介
吴永洪(1989-),男,在读研究生,主要研究方向为机器学习、电力负荷预测;通讯作者:张智斌(1965-),男,副教授,主要研究方向为基于网络的计算机软件技术、工业控制技术。
